Most organisations investing seriously in diversity, equity and inclusion have put real effort into the visible parts of the problem: inclusive job descriptions, diverse interview panels, anonymised screening, and graduate partnerships with universities that serve underrepresented groups. These efforts matter. But on their own, they’re not enough, because they don’t address a less visible problem that sits upstream of all of them.

The talent intelligence most organisations rely on to understand their market, benchmark their employer brand, and shape their attraction strategies, is systematically underrepresenting diverse talent. Not because of bias in how the data is analysed, but because of where the data comes from in the first place.

If you’re measuring the market using data sources that over-index on certain kinds of candidates, your DEI strategy will always be playing catch-up. You’ll be trying to attract diverse talent while your intelligence layer tells you nothing reliable about who they are, what they want, and where the gaps in your current approach actually lie.

Where the problem starts

Traditional labour market data draws from three main sources: job postings, professional network profiles (mainly LinkedIn), and employer-commissioned surveys. Each of these has a diversity bias built into how it works.

Job postings reflect what employers have historically asked for. When those requirements have traditionally favoured candidates from certain educational backgrounds or career trajectories, the data derived from those postings simply encodes and repeats those same preferences. Analysing market-wide skills demand is not a neutral exercise if the demands themselves reflect historical inequality.

LinkedIn profile data over-indexes on candidates who are active on the platform, comfortable with professional networking norms, and have the time to maintain a curated online presence. Research consistently shows that this skews towards higher socioeconomic backgrounds, towards men in technical roles, and towards candidates who are already relatively visible in the professional market. The talent that is most underrepresented in traditional pipelines tends to be equally underrepresented in LinkedIn's data.

Employer-commissioned surveys face a different issue: response bias. Who responds to workplace surveys, and how honestly they respond, is closely linked to demographic factors. Employees from underrepresented groups often respond more cautiously, aware of the potential consequences of candour. The result is survey data that does not accurately reflect the real experiences or priorities of the most diverse parts of your workforce.

What this means for your organisation

If your talent intelligence is built on these data sources, there are specific consequences that are likely to be showing up in your data already, even if the root cause is not obvious.

Your benchmarks for competitive pay and benefits are probably calibrated against a candidate population that doesn’t fully reflect the diversity of talent you want to attract. Your EVP is probably shaped around priorities that reflect the stated preferences of your most survey-active employees, who may not represent your most diverse colleagues. And your understanding of what actually drives attraction among neurodivergent candidates, disabled job seekers, or candidates from lower socioeconomic backgrounds is probably weak, because those groups are the least well-served by the methods you are using to understand them.

The result is that DEI efforts tend to focus on adjusting your process at the application and selection stage, rather than on the earlier and arguably more important question of whether your employer proposition is genuinely visible and appealing to diverse talent in the first place.

The insight gap most DEI strategies miss

There is a specific gap that most DEI strategies have not yet addressed: the difference between what diverse candidates say matters to them, and what their behaviour shows they’re actually looking for. This gap exists for all candidates, but it’s particularly pronounced for underrepresented groups, where survey responses can be shaped by an awareness of what employers want to hear, and a reasonable caution about disclosing certain priorities.

Behavioural data sidesteps this entirely. When neurodivergent candidates consistently apply specific filters around predictable working patterns and sensory-friendly environments, that is a revealed preference. When candidates with disabilities filter for office accessibility and remote-first arrangements before anything else, that is a revealed preference too. These are insights that no survey would reliably surface, but they have direct implications for how employers should design and communicate their proposition to reach these talent pools effectively.

For LGBTQ+ candidates, the same principle applies. Their search behaviour consistently shows that flexible working and mental health support are not secondary considerations:they’re often the deciding factors.

From better data to better outcomes

The organisations achieving real improvements in pipeline diversity are not simply doing more of the same DEI activities. They are building their strategies on better intelligence about what diverse talent actually wants, and where the genuine gaps in their current offer lie.

The results speak for themselves. Employers such as Amplifi are using Flexa's behavioural data to inform their DEI strategy, and have seen aggregate diversity across their talent pools increase by 65%. Airbus, a global leader in aeronautics, increased gender diversity in their pipeline by 91% in three months, not through targeted campaigns, but by making its working model visible and discoverable to talent that had not previously considered it.

Improving diversity outcomes in talent attraction is a data problem, not a creativity problem. And, put simply: the data that most organisations are using to make these decisions is the wrong data.

Flexa's platform connects employers with a verified diverse talent pool of 4.5 million users, with insight into what different groups prioritise and how your employer proposition compares. Learn how to interpret your data to better understand and improve diverse talent attraction.